| dc.description.abstract | A variety of products with various categories are currently being marketed online. E commerce users have many choices in meeting their needs or desires. One of the factors
that e-commerce users concern is the product reviews given by other users who have
purchased the product. Online sellers can monitor the quality of their service and
products through product reviews to take action. However, the difficulty of evaluating
the performance of an online store is a big challenge for online sellers because it must
be done manually and requires a prolonged time and good concentration to deliver the
appropriate information. This study aims to assist online sellers in conducting aspect based sentiment analysis on product reviews by combining the Convolutional Neural
Network algorithm with the Long-Short Term Memory algorithm or CNN-LSTM. This
study uses 7,500 product review data that went through preprocessing stages that
consist of case folding, data cleaning, data normalization, stemming, and tokenization,
followed by a word embedding stage using the fastText library. The aspect model built
in this study has been well used to classify
six aspect categories of Indonesian product reviews, namely accuracy, quality, service,
packaging, price, and delivery, which produces an average accuracy of 93.58%; and
the sentiment model built is also able to analyze sentiment from classified review texts,
with an average sentiment model accuracy of 91.97%. | en_US |
